I have got my master’s degree at the Computer Vision Institute, Shenzhen University, under the supervision of Prof.Feng Liu and Prof.Linlin Shen. Besides, I was an intern at Jarvis Lab, Tencent. My major research interests include image & video generation and system security.
🔥 News
- 2024.01: 🎉 I am invited as a reviewer for ECCV’2024!
- 2023.11: 🎉 I am invited as a reviewer for CVPR’2024!
- 2023.09: 🎉 Our paper (Dynamically Masked Discriminator for GANs) is accepted by NeurIPS’2023!
- 2023.07: 🎉 Our paper (BoxDiff: Text-to-Image Synthesis with Training-Free Box-Constrained Diffusion) is accepted by ICCV’2023!
- 2023.02: 🎉 Our paper (AdaptiveMix: Robust Feature Representation via Shrinking Feature Space.) is accepted by CVPR’2023!
- 2023.02: 🎉 Our paper (NewsNet: A Novel Dataset for Hierarchical Temporal Segmentation.) is accepted by CVPR’2023!
- 2022.09: I am awarded the China National Scholarship.
- 2022.08: 🎉 Our paper (Decoupled Mixup for Out-of-Distribution Visual Recognition) is invited as a regular paper in the ECCV’2022 Workshop!
- 2022.08: Our method (Decoupled Mixup) reaches to the 4th/40 in Out-of-Distribution Visual Recognition ECCV’2022 NICO Challenge.！
- 2022.07: 🎉 Our paper (Effective Presentation Attack Detection Driven by Face Related Task) is accepted by ECCV’2022!
- 2022.06: 🎉 Our paper (A Multi-task Network with Weight Decay Skip Connection Training for Anomaly Detection in Retinal Fundus Images) is accepted by MICCAI’2022!
📝 Publications
NIPS 2023
Dynamically Masked Discriminator for GANs.
Wentian Zhang, Haozhe Liu, Bing Li, Jinheng Xie, Yawen Huang, Yuexiang Li, Yefeng Zheng, Bernard Ghanem
- In the GAN training, We observe that the discriminator model, trained on historically generated data, often slows down its adaptation to the changes in the new arrival generated data, which accordingly decreases the quality of generated results. We propose a new discriminator, which automatically detects its retardation and then dynamically masks its features, such that the discriminator can adaptively learn the temporally-vary distribution of generated data. Code
CVPR 2023
AdaptiveMix: Robust Feature Representation via Shrinking Feature Space.
Haozhe Liu†, Wentian Zhang†, Bing Li, Haoqian Wu, Nanjun He, Yawen Huang, Yuexiang Li, Bernard Ghanem, Yefeng Zheng(† Equal Contribution)
- Training GANs is difficult since the training distribution is dynamic for the discriminator, leading to unstable image representation. We address the problem of training GANs from a novel perspective, i.e., robust image classification. We propose a simple yet effective module, namely AdaptivelyMix, for GANs, which shrinks the regions of training data in the image representation space of the discriminator. Code
ECCVW 2022
Decoupled Mixup for Out-of-Distribution Visual Recognition.
Haozhe Liu†, Wentian Zhang†, Jinheng Xie†, Haoqian Wu, Bing Li, Ziqi Zhang, Yuexiang Li, Yawen Huang, Bernard Ghanem, Yefeng Zheng(† Equal Contribution)
- We propose a novel ”Decoupled-Mixup” method to train CNN models for OOD visual recognition. Different from previous work combining pairs of images homogeneously, our method decouples each image into discriminative and noise-prone regions, and then heterogeneously combine these regions of image pairs to train CNN models. Code
ECCV 2022
Effective Presentation Attack Detection Driven by Face Related Task
Wentian Zhang;, Haozhe Liu;, Feng Liu, Raghavendra Ramachandra, Christoph Busch
- Unlike face PAD task, other face related tasks trained by huge amount of real faces (e.g. face recognition and attribute editing) can be effectively adopted into different application scenarios. Inspired by this, we propose to trade position of PAD and face related work in a face system and apply the free acquired prior knowledge from face related tasks to solve face PAD, so as to improve the generalization ability in detecting PAs. Code
MICCAI 2022
Wentian Zhang;, Xu Sun;, Yuexiang Li;, Haozhe Liu, Nanjun He, Feng Liu, Yefeng Zheng
- We propose a weight decay training strategy to progressively mute the skip connections of U-Net, which effectively adapts U-shape network to anomaly detection task. Furthermore, we formulate histograms of oriented gradients (HOG) prediction to encourage the framework to deeply exploit contextual information from fundus images. Code
Selected Publication List
- Zhang, W.,Liu, H., Li, B., Xie, J., Huang, Y., Li, Y., Zheng, Y., & Ghanem, B. (2023). Dynamically Masked Discriminator for GANs. (NeurIPS’2023).
- Zhang, W., Liu, H., Liu, F., Ramachandra, R., & Busch, C. (2022). Effective Presentation Attack Detection Driven by Face Related Task. (ECCV’2022).
- Zhang, W., Sun, X., Li, Y., Liu, H., He, N., Liu, F., & Zheng, Y. (2022). A Multi-task Network with Weight Decay Skip Connection Training for Anomaly Detection in Retinal Fundus Images. (MICCAI’2022).
- Liu, H.*, Zhang, W.*, Li, B., Wu, H., He, N., Huang, Y., Li, Y., Ghanem, B., & Zheng, Y. (2023). AdaptiveMix: Robust Feature Representation via Shrinking Feature Space. (CVPR’2023). (* Equal Contribution)
- Kong, Z.*, Zhang, W.*, Liu, F., Luo, W., Liu, H., Shen, L., Raghavendra, R. (2023). Taming Self-Supervised Learning for Presentation Attack Detection: De-Folding and De-Mixing. IEEE T-NNLS (* Equal Contribution)
- Liu, H.*, Zhang, W.*, Xie J.*, Wu, H., Li, B., Zhang, Z., Li, Y., Huang, Y., Ghanem, B., & Zheng, Y. (2022). Decoupled Mixup for Out-of-Distribution Visual Recognition. (ECCVW’2022). (* Equal Contribution)
- Wu, H., Chen, K., Liu, H., Zhuge, M., Li, B., Qiao, R., Shu, X., Gan, B., Xu, L., Ren, B., Xu, M., Zhang, W., Ramachandra, R., Lin, C., & Ghanem, B. (2023) NewsNet: A Novel Dataset for Hierarchical Temporal Segmentation. (CVPR’2023).
- Xie, J., Li, Y., Huang, Y., Liu, H., Zhang, W., Zheng, Y., & Shou, M. Z. (2023). BoxDiff: Text-to-Image Synthesis with Training-Free Box-Constrained Diffusion. (ICCV’2023).
🎖 Honors and Awards
- 2022 China National Scholarship (Rate<0.02%)
- 2021 Excellent Academic Scholarship, First Class
- 2020 Excellent Academic Scholarship, First Class
- 2018 National University Big Data Application Innovation Competition, First Place
📖 Research Experience
AI Initiative (KAUST)
Visiting student supervised by Dr. Bing Li. Closely cooperate with Haozhe Liu
- Proposed an ownership protection method for generative models (diffusion models and GAN models).
- Proposed a dynamically masked discriminator for generative adversarial networks, which is accepted by NeurIPS’2023.
- Proposed a robust adversarial learning method by shrinking feature space in the training phase, which is accepted by CVPR’2023.
- Participated in establishing a novel dataset for hierarchical temporal segmentation, which is accepted by CVPR’2023.
Jarvis Lab (Tencent)
Internship supervised by Mentor: Dr. Xu Sun & Dr. Yuexiang Li and Director: Dr. Yefeng Zheng
- Proposed a weight decay strategy to progressively mute the skip connections of U-Net for anomaly detection task, which is accepted by MICCAI’2022.
- Participate to NICO Challenge (ECCV’2022 Workshop), our team reach to 5th/40 in both tracks at Phase I, and 4th in Track 2 at Final Phase.
Norwegian Biometrics Laboratory (NTNU)
Collaborating with Prof. Raghavendra Ramachandra
- Proposed a face presentation attack detector based on the taskonomy features, which is accepted by ECCV’2022.
Shenzhen Institute of Artificial Intelligence and Robotics for Society (CUHK)
Visiting student supervised by Prof. David Zhang
- Participated in collecting a multi-modal biometric dataset, which contains face, fingerprint and palmprint samples from 10k subjects.
- Proposed to apply a 3D convolution network to extract palmprint features which can be further encoded for recognition.
Computer Vision Institute (Shenzhen University)
M.S. supervised by Prof. Feng Liu and Prof. Linlin Shen
- Proposed a uniform representation learning method for OCT-based Fingerprint anti-spoofing and Recognition, which is accepted by Pattern Recognition.
- Proposed a minutiae extraction model with a fusion-attention mechanism for multi-layered OCT fingerprints.
- Proposed to establish a one-class framework for OCT-based PAD. This work is accepted by IEEE TIP.